OpenAI has introduced a Computer History feature for the ChatGPT desktop application on macOS that records user activity to improve AI responses [1].
The tool represents a shift toward deeper system integration, allowing an AI to understand a user's actual behavior rather than relying solely on manual prompts. This capability enables the assistant to reference a user's recent workflow to provide more relevant and continuous answers [1].
Computer History is an optional feature that logs clicks, keystrokes, and window titles [2]. These data points are used to build a timeline of activity that the model can reference when answering questions [3]. According to the OpenAI blog, the feature is designed to help the model provide more relevant responses [1].
Sam Altman, CEO of OpenAI, said the company understands the privacy implications and has made the feature opt-in [2]. However, reports on the default status of the tool vary. ZDNet said the feature is disabled by default and must be turned on by the user [2], while Tom's Guide said it is enabled by default upon installation [4].
Technical details regarding the data indicate that the feature can create a timeline spanning up to 30 days of activity [1]. The Mac application with this feature enabled is approximately 150 MB in size [3].
There are contradictions regarding how the data is handled. Digital Trends said data is stored locally on the Mac and encrypted before being sent to OpenAI servers [1]. Conversely, The Verge said Computer History streams activity data directly to the cloud for real-time processing [5].
OpenAI rolled out the feature in March 2026 [1]. The company said the goal is to give the assistant context about a user's recent workflow to improve the continuity of its responses [1].
“Computer History is an optional feature that records your activity on your Mac to help the model provide more relevant responses.”
The introduction of Computer History marks a transition from conversational AI to an agentic model that monitors system-level telemetry. By tracking keystrokes and window titles, OpenAI is moving toward a 'world model' of the user's digital environment. This creates a tension between utility and privacy, as the ability to provide seamless context requires the collection of highly sensitive behavioral data, raising questions about the security of local versus cloud-based storage.
